AI Doesn’t Feel Empathy. Can It Help Managers Communicate It?

“The document is incomplete. Fix it by the end of today and tell me when you are done.”

“I have reviewed the document and a few elements still need to be added before it is complete. Could you include them by the end of today and update me when it is ready? If anything is getting in the way, let me know so that we can decide quickly how to proceed.”

The task is essentially the same. The experience of receiving it may be very different.

The first message can sound rushed, closed or reproachful. The second preserves the deadline and the employee’s responsibility, while adding context, an opening for dialogue and a way to surface obstacles.

This does not necessarily mean that the second manager is more empathetic. They may simply have asked an artificial intelligence system to improve the message.

Yet the organisational effect can still be real: less energy spent interpreting tone, less resistance and greater clarity about what needs to happen.

That leads to a more useful question than “can AI write well?”

Can it reduce friction in working relationships without turning managerial communication into artificial politeness?

A correct request can still create the wrong response

In organisations, tone is not decoration added to content.

An email assigns a task, but it also communicates urgency, trust, availability, hierarchy and room for autonomy. A technically correct sentence can create irritation or misunderstanding. A request softened too much can obscure priorities and accountability.

This is particularly visible when workloads are high, the relationship is already strained, or a message arrives without the support of voice, expression and shared context.

Part of a manager’s job is to hold two needs together: making things happen and protecting the quality of the relationship through which they will happen.

AI can operate in the gap between intention and possible impact. It can surface wording that may sound abrupt, expectations that have not been made explicit, ambiguity or excessive politeness that makes a message less credible.

But it does not truly know the history between two people. And it will not live with the consequences of the communication.

AI as a mirror, not a mask

There is an important difference between using AI to see your own message more clearly and using it to sound like a different person.

As a mirror, it can surface a blind spot: “this sentence may be read as a rebuke”; “you have not explained why the deadline matters”; “you appear to invite a response, but leave no real space for one”.

As a mask, it produces flawless warmth that the manager’s daily behaviour does not support.

In the first case, AI supports awareness and choice. In the second, it creates distance between language and experience.

Research supports this distinction. A study involving 1,100 working professionals found that AI-assisted writing may appear professional, while more extensive AI assistance used by a manager can weaken dimensions of perceived trustworthiness, including sincerity, caring, competence and confidence about who really authored the text.

The issue, then, is not simply whether the message is good.

It is whether the recipient can still recognise an accountable person behind it.

The more sensitive the message, the less good wording is enough

Not every communication carries the same relational weight.

For an operational request, AI can be extremely helpful in removing ambiguity, organising information and checking tone.

With an apology, difficult feedback or conflict, the risk is different. Research on AI-mediated apologies in a multilingual workplace context found that more visible AI mediation could reduce perceived authenticity and willingness to forgive. Limited assistance did not have the same negative effect.

This is consistent with everyday experience. When something has hurt us, we do not only look for correct words. We look for evidence that the other person understands the impact and is willing to take responsibility for it.

AI can help prepare for a conversation. It cannot resolve the conversation a manager is avoiding.

When a relationship requires presence, a perfect email may become an elegant form of absence.

When the problem is not intention, but language

Writing in a language other than your own adds another layer.

It does not mean that people working in a second language are less thoughtful or less empathetic. It means that the distance between what they intend and what the reader perceives may become wider.

Urgency, disagreement, authority, availability and courtesy are not expressed in the same way across languages and cultures. A literal translation may be too direct. A highly formal text may feel distant. A sentence designed to be polite may become vague.

AI can play a valuable role here: not simply translating words, but helping to translate relational intention.

It can help explore questions such as:

• could this sentence sound more abrupt to a native reader than I intend?
• does my request make priorities, accountability and room for dialogue clear?
• is there an alternative that preserves my voice rather than imitating a tone that is not mine?

Caution still matters. A recent preregistered study of professional writing among Japanese and American participants suggests that AI assistance can shift or flatten culturally distinctive communication patterns. There is no universally correct version of professional courtesy.

The objective should not be to make everyone sound like the same polished corporate English speaker. It should be to reduce misunderstanding while preserving voice, context and mutual respect.

A learning tool or an outsourced capability?

The same action — asking AI to review a message — can produce two opposite outcomes.

If someone copies the proposed version without comparing it with their own, they gain a better text but not necessarily a new capability.

If they observe what changed, ask why and formulate another option, AI becomes a practice environment.

This is particularly relevant for early-career professionals, people working in a second language and first-time managers. Not because these groups are less capable, but because they are still building references for situations they have not encountered many times before.

A qualitative study with 17 managers highlighted the value participants placed on low-risk conversational practice, iterative feedback and the opportunity to work on the balance between empathy and assertiveness. They saw AI as an addition to human training and mentoring, rather than a replacement.

This distinction connects with a wider challenge: if AI performs the work through which people used to learn, organisations must design new learning opportunities deliberately. We explore that question further in our article on AI and entry-level work.

Do not simply ask it to “make this more empathetic”

A generic instruction often produces language that is correct but impersonal.

It is more useful to ask AI to support the thinking before it rewrites the text.

For example:

• As a mirror: “Identify phrases that may sound abrupt, ambiguous or passive-aggressive. Do not rewrite the message yet; explain the possible interpretations.”
• To preserve clarity: “The request must remain firm and the deadline is not negotiable. Offer two formulations that explain the reason without weakening the recipient’s accountability.”
• To learn: “Compare my version with yours. Highlight three changes and explain what different response each change might encourage.”
• Across languages: “Translate this into English while preserving my direct style. Flag phrases that may sound abrupt to a native reader and suggest alternatives, explaining the difference.”

This is not about predicting with certainty how a person will react. No system can do that.

It is about considering more possible readings before pressing “send”.

Three questions before sending

Research on AI-assisted workplace communication points to two possibilities: messages can become more civil and comprehensible, but people can also become dependent, less authentic and less skilled. A proposed policy for AI-mediated workplace communication captures the underlying principle well: the human should remain in the pilot’s seat.

A manager can operationalise that principle with three questions.

• Intention: what does the recipient need to understand, decide or do?
• Impact: how might they read this message, given the relationship, context, language and moment?
• Responsibility: do I recognise these words as mine, and am I prepared to stand behind them in a conversation?

For anyone still developing the capability, I would add a fourth:

Learning: what did AI change, and what can I carry into my next message without needing its help?

When to use it — and when to stop

AI is particularly useful for:

• making operational requests, priorities and deadlines clearer;
• reducing ambiguity and unnecessary length;
• checking for wording that may unintentionally sound abrupt;
• adapting a text for another language or an international audience;
• preparing alternatives before a difficult conversation;
• practising clear and respectful feedback.

It requires much greater caution when the message contains an apology, meaningful recognition, conflict, performance feedback, a career decision, a disciplinary matter or something personal.

In those situations, the question is not only “how should I write this?” It is also “should this communication be written at all?” and “which part must I take responsibility for directly?”

More human communication does not emerge automatically from kinder words

AI does not feel empathy. It does not know what it means to work with a particular person, remember a conflict, recognise vulnerability or decide to repair a relationship.

It can, however, reveal something we miss when we are rushed.

It can show us that a correct request sounds like a rebuke. That a precise translation does not carry our intention. That we have demanded accountability without offering enough context. Or that we are using deeply empathetic language to avoid an authentic conversation.

The value is not in making a manager appear kinder.

It is in helping them become more aware of the effect of their words without surrendering responsibility for the relationship.

Because a good message is not simply one that sounds right.

It is one in which clarity, intention and behaviour continue to describe the same person.